ACL 2022 Papers

Existing reference-free metrics have obvious limitations for evaluating controlled text generation models. Unsupervised metrics can only provide a task-agnostic evaluation result which correlates weakly with human judgments, whereas supervised ones may overfit task-specific data with poor generalization ability to other datasets. In this paper, we propose an unsupervised reference-free metric called CTRLEval, which evaluates controlled text generation from different aspects by formulating each aspect into multiple text infilling tasks. On top of these tasks, the metric assembles the generation probabilities from a pre-trained language model without any model training. Experimental results show that our metric has higher correlations with human judgments than other baselines, while obtaining better generalization of evaluating generated texts from different models and with different qualities.

Recently, Bert-based models have dominated the research of Chinese spelling correction (CSC). These methods have two limitations: (1) they have poor performance on multi-typo texts. In such texts, the context of each typo contains at least one misspelled character, which brings noise information. Such noisy context leads to the declining performance on multi-typo texts. (2) they tend to overcorrect valid expressions to more frequent expressions due to the masked token recovering task of Bert. We attempt to address these limitations in this paper. To make our model robust to contextual noise brought by typos, our approach first constructs a noisy context for each training sample. Then the correction model is forced to yield similar outputs based on the noisy and original contexts. Moreover, to address the overcorrection problem, copy mechanism is incorporated to encourage our model to prefer to choose the input character when the miscorrected and input character are both valid according to the given context. Experiments are conducted on widely used benchmarks. Our model achieves superior performance against state-of-the-art methods by a remarkable gain.

Humans are able to perceive, understand and reason about causal events. Developing models with similar physical and causal understanding capabilities is a long-standing goal of artificial intelligence. As a step towards this direction, we introduce CRAFT, a new video question answering dataset that requires causal reasoning about physical forces and object interactions. It contains 58K video and question pairs that are generated from 10K videos from 20 different virtual environments, containing various objects in motion that interact with each other and the scene. Two question categories in CRAFT include previously studied descriptive and counterfactual questions. Additionally, inspired by the Force Dynamics Theory in cognitive linguistics, we introduce a new causal question category that involves understanding the causal interactions between objects through notions like cause, enable, and prevent. Our results show that even though the questions in CRAFT are easy for humans, the tested baseline models, including existing state-of-the-art methods, do not yet deal with the challenges posed in our benchmark.

Multi-hop question generation focuses on generating complex questions that require reasoning over multiple pieces of information of the input passage. Current models with state-of-the-art performance have been able to generate the correct questions corresponding to the answers. However, most models can not ensure the complexity of generated questions, so they may generate shallow questions that can be answered without multi-hop reasoning. To address this challenge, we propose the CQG, which is a simple and effective controlled framework. CQG employs a simple method to generate the multi-hop questions that contain key entities in multi-hop reasoning chains, which ensure the complexity and quality of the questions. In addition, we introduce a novel controlled Transformer-based decoder to guarantee that key entities appear in the questions. Experiment results show that our model greatly improves performance, which also outperforms the state-of-the-art model about 25% by 5 BLEU points on HotpotQA.

Named Entity Recognition (NER) in Few-Shot setting is imperative for entity tagging in low resource domains. Existing approaches only learn class-specific semantic features and intermediate representations from source domains. This affects generalizability to unseen target domains, resulting in suboptimal performances. To this end, we present CONTaiNER, a novel contrastive learning technique that optimizes the inter-token distribution distance for Few-Shot NER. Instead of optimizing class-specific attributes, CONTaiNER optimizes a generalized objective of differentiating between token categories based on their Gaussian-distributed embeddings. This effectively alleviates overfitting issues originating from training domains. Our experiments in several traditional test domains (OntoNotes, CoNLL’03, WNUT ‘17, GUM) and a new large scale Few-Shot NER dataset (Few-NERD) demonstrate that on average, CONTaiNER outperforms previous methods by 3%-13% absolute F1 points while showing consistent performance trends, even in challenging scenarios where previous approaches could not achieve appreciable performance.

Supervised learning has traditionally focused on inductive learning by observing labeled examples of a task. In contrast, a hallmark of human intelligence is the ability to learn new concepts purely from language. Here, we explore training zero-shot classifiers for structured data purely from language. For this, we introduce CLUES, a benchmark for Classifier Learning Using natural language ExplanationS, consisting of a range of classification tasks over structured data along with natural language supervision in the form of explanations. CLUES consists of 36 real-world and 144 synthetic classification tasks. It contains crowdsourced explanations describing real-world tasks from multiple teachers and programmatically generated explanations for the synthetic tasks. To model the influence of explanations in classifying an example, we develop ExEnt, an entailment-based model that learns classifiers using explanations. ExEnt generalizes up to 18% better (relative) on novel tasks than a baseline that does not use explanations. We delineate key challenges for automated learning from explanations, addressing which can lead to progress on CLUES in the future. Code and datasets are available at: https://clues-benchmark.github.io.

CLIP has shown a remarkable zero-shot capability on a wide range of vision tasks. Previously, CLIP is only regarded as a powerful visual encoder. However, after being pre-trained by language supervision from a large amount of image-caption pairs, CLIP itself should also have acquired some few-shot abilities for vision-language tasks. In this work, we empirically show that CLIP can be a strong vision-language few-shot learner by leveraging the power of language. We first evaluate CLIP’s zero-shot performance on a typical visual question answering task and demonstrate a zero-shot cross-modality transfer capability of CLIP on the visual entailment task. Then we propose a parameter-efficient fine-tuning strategy to boost the few-shot performance on the vqa task. We achieve competitive zero/few-shot results on the visual question answering and visual entailment tasks without introducing any additional pre-training procedure.

This paper addresses the problem of dialogue reasoning with contextualized commonsense inference. We curate CICERO, a dataset of dyadic conversations with five types of utterance-level reasoning-based inferences: cause, subsequent event, prerequisite, motivation, and emotional reaction. The dataset contains 53,105 of such inferences from 5,672 dialogues. We use this dataset to solve relevant generative and discriminative tasks: generation of cause and subsequent event; generation of prerequisite, motivation, and listener’s emotional reaction; and selection of plausible alternatives. Our results ascertain the value of such dialogue-centric commonsense knowledge datasets. It is our hope that CICERO will open new research avenues into commonsense-based dialogue reasoning.

Artificial Intelligence (AI), along with the recent progress in biomedical language understanding, is gradually offering great promise for medical practice. With the development of biomedical language understanding benchmarks, AI applications are widely used in the medical field. However, most benchmarks are limited to English, which makes it challenging to replicate many of the successes in English for other languages. To facilitate research in this direction, we collect real-world biomedical data and present the first Chinese Biomedical Language Understanding Evaluation (CBLUE) benchmark: a collection of natural language understanding tasks including named entity recognition, information extraction, clinical diagnosis normalization, single-sentence/sentence-pair classification, and an associated online platform for model evaluation, comparison, and analysis. To establish evaluation on these tasks, we report empirical results with the current 11 pre-trained Chinese models, and experimental results show that state-of-the-art neural models perform by far worse than the human ceiling.

We introduce CARETS, a systematic test suite to measure consistency and robustness of modern VQA models through a series of six fine-grained capability tests. In contrast to existing VQA test sets, CARETS features balanced question generation to create pairs of instances to test models, with each pair focusing on a specific capability such as rephrasing, logical symmetry or image obfuscation. We evaluate six modern VQA systems on CARETS and identify several actionable weaknesses in model comprehension, especially with concepts such as negation, disjunction, or hypernym invariance. Interestingly, even the most sophisticated models are sensitive to aspects such as swapping the order of terms in a conjunction or varying the number of answer choices mentioned in the question. We release CARETS to be used as an extensible tool for evaluating multi-modal model robustness.

Model ensemble is a popular approach to produce a low-variance and well-generalized model. However, it induces large memory and inference costs, which is often not affordable for real-world deployment. Existing work has resorted to sharing weights among models. However, when increasing the proportion of the shared weights, the resulting models tend to be similar, and the benefits of using model ensemble diminish. To retain ensemble benefits while maintaining a low memory cost, we propose a consistency-regularized ensemble learning approach based on perturbed models, named CAMERO. Specifically, we share the weights of bottom layers across all models and apply different perturbations to the hidden representations for different models, which can effectively promote the model diversity. Meanwhile, we apply a prediction consistency regularizer across the perturbed models to control the variance due to the model diversity. Our experiments using large language models demonstrate that CAMERO significantly improves the generalization performance of the ensemble model. Specifically, CAMERO outperforms the standard ensemble of 8 BERT-base models on the GLUE benchmark by 0.7 with a significantly smaller model size (114.2M vs. 880.6M).

Knowledge graphs store a large number of factual triples while they are still incomplete, inevitably. The previous knowledge graph completion (KGC) models predict missing links between entities merely relying on fact-view data, ignoring the valuable commonsense knowledge. The previous knowledge graph embedding (KGE) techniques suffer from invalid negative sampling and the uncertainty of fact-view link prediction, limiting KGC’s performance. To address the above challenges, we propose a novel and scalable Commonsense-Aware Knowledge Embedding (CAKE) framework to automatically extract commonsense from factual triples with entity concepts. The generated commonsense augments effective self-supervision to facilitate both high-quality negative sampling (NS) and joint commonsense and fact-view link prediction. Experimental results on the KGC task demonstrate that assembling our framework could enhance the performance of the original KGE models, and the proposed commonsense-aware NS module is superior to other NS techniques. Besides, our proposed framework could be easily adaptive to various KGE models and explain the predicted results.

Existing commonsense knowledge bases often organize tuples in an isolated manner, which is deficient for commonsense conversational models to plan the next steps. To fill the gap, we curate a large-scale multi-turn human-written conversation corpus, and create the first Chinese commonsense conversation knowledge graph which incorporates both social commonsense knowledge and dialog flow information. To show the potential of our graph, we develop a graph-conversation matching approach, and benchmark two graph-grounded conversational tasks. All the resources in this work will be released to foster future research.

We consider the problem of pretraining a two-stage open-domain question answering (QA) system (retriever + reader) with strong transfer capabilities. The key challenge is how to construct a large amount of high-quality question-answer-context triplets without task-specific annotations. Specifically, the triplets should align well with downstream tasks by: (i) covering a wide range of domains (for open-domain applications), (ii) linking a question to its semantically relevant context with supporting evidence (for training the retriever), and (iii) identifying the correct answer in the context (for training the reader). Previous pretraining approaches generally fall short of one or more of these requirements. In this work, we automatically construct a large-scale corpus that meets all three criteria by consulting millions of references cited within Wikipedia. The well-aligned pretraining signals benefit both the retriever and the reader significantly. Our pretrained retriever leads to 2%-10% absolute gains in top-20 accuracy. And with our pretrained reader, the entire system improves by up to 4% in exact match.

Pretrained language models such as BERT have achieved remarkable success in several NLP tasks. With the wide adoption of BERT in real-world applications, researchers begin to investigate the implicit biases encoded in the BERT. In this paper, we assess the implicit stock market preferences in BERT and its finance domain-specific model FinBERT. We find some interesting patterns. For example, the language models are overall more positive towards the stock market, but there are significant differences in preferences between a pair of industry sectors, or even within a sector. Given the prevalence of NLP models in financial decision making systems, this work raises the awareness of their potential implicit preferences in the stock markets. Awareness of such problems can help practitioners improve robustness and accountability of their financial NLP pipelines .

Text-to-SQL parsers map natural language questions to programs that are executable over tables to generate answers, and are typically evaluated on large-scale datasets like Spider (Yu et al., 2018). We argue that existing benchmarks fail to capture a certain out-of-domain generalization problem that is of significant practical importance: matching domain specific phrases to composite operation over columns. To study this problem, we first propose a synthetic dataset along with a re-purposed train/test split of the Squall dataset (Shi et al., 2020) as new benchmarks to quantify domain generalization over column operations, and find existing state-of-the-art parsers struggle in these benchmarks. We propose to address this problem by incorporating prior domain knowledge by preprocessing table schemas, and design a method that consists of two components: schema expansion and schema pruning. This method can be easily applied to multiple existing base parsers, and we show that it significantly outperforms baseline parsers on this domain generalization problem, boosting the underlying parsers’ overall performance by up to 13.8% relative accuracy gain (5.1% absolute) on the new Squall data split.

Back-translation is a critical component of Unsupervised Neural Machine Translation (UNMT), which generates pseudo parallel data from target monolingual data. A UNMT model is trained on the pseudo parallel data with translated source, and translates natural source sentences in inference. The source discrepancy between training and inference hinders the translation performance of UNMT models. By carefully designing experiments, we identify two representative characteristics of the data gap in source: (1) style gap (i.e., translated vs. natural text style) that leads to poor generalization capability; (2) content gap that induces the model to produce hallucination content biased towards the target language. To narrow the data gap, we propose an online self-training approach, which simultaneously uses the pseudo parallel data {natural source, translated target} to mimic the inference scenario. Experimental results on several widely-used language pairs show that our approach outperforms two strong baselines (XLM and MASS) by remedying the style and content gaps.

In recent years, large-scale pre-trained language models (PLMs) have made extraordinary progress in most NLP tasks. But, in the unsupervised POS tagging task, works utilizing PLMs are few and fail to achieve state-of-the-art (SOTA) performance. The recent SOTA performance is yielded by a Guassian HMM variant proposed by He et al. (2018). However, as a generative model, HMM makes very strong independence assumptions, making it very challenging to incorporate contexualized word representations from PLMs. In this work, we for the first time propose a neural conditional random field autoencoder (CRF-AE) model for unsupervised POS tagging. The discriminative encoder of CRF-AE can straightforwardly incorporate ELMo word representations. Moreover, inspired by feature-rich HMM, we reintroduce hand-crafted features into the decoder of CRF-AE. Finally, experiments clearly show that our model outperforms previous state-of-the-art models by a large margin on Penn Treebank and multilingual Universal Dependencies treebank v2.0.

While multilingual training is now an essential ingredient in machine translation (MT) systems, recent work has demonstrated that it has different effects in different multilingual settings, such as many-to-one, one-to-many, and many-to-many learning. These training settings expose the encoder and the decoder in a machine translation model with different data distributions. In this paper, we examine how different varieties of multilingual training contribute to learning these two components of the MT model. Specifically, we compare bilingual models with encoders and/or decoders initialized by multilingual training. We show that multilingual training is beneficial to encoders in general, while it only benefits decoders for low-resource languages (LRLs). We further find the important attention heads for each language pair and compare their correlations during inference. Our analysis sheds light on how multilingual translation models work and also enables us to propose methods to improve performance by training with highly related languages. Our many-to-one models for high-resource languages and one-to-many models for LRL outperform the best results reported by Aharoni et al. (2019).